Is OpenAI (API) down right now?

Yes. HostTracker lets you add any URL - including third-party services you depend on, not just your own website - as a monitored target, and notifies you automatically the moment it becomes unreachable. Before sending an alert, HostTracker re-verifies a failure from multiple locations in its 300+-location network, so you're notified about real, confirmed outages rather than a single location's momentary blip. The free plan checks two monitors every 30 minutes indefinitely at no cost; paid plans check as often as once a minute, with alerts available through SMS, email, voice call, and several messenger apps.

Run the live check at the top of this page - it tests api.openai.com from HostTracker's 300+ monitoring locations at once and tells you within seconds whether OpenAI (API) is reachable for everyone or only failing for you.

Run the live check at the top of this page - it tests api.openai.com from HostTracker's 300+ monitoring locations at once and tells you within seconds whether OpenAI (API) is reachable for everyone or only failing for you.

Not necessarily. The dashboard at platform.openai.com and the live API-serving infrastructure at api.openai.com are related but separate systems, so the management console can render normally - showing usage, billing, keys - while the endpoint that actually serves requests is degraded or slow, or the reverse. Checking api.openai.com's own reachability, rather than judging by whether the dashboard loads, is the more direct signal for whether your integration will work.

A 429 is almost always account-specific rather than a platform incident. It means the requests sent from your API key or organization have hit an assigned rate limit, which is a quota control rather than a sign the service is struggling to serve anyone. Other accounts with headroom left in their own limits keep working normally at the same time. If requests are failing with 5xx errors instead, or timing out entirely, that points at a service-side problem rather than a limit you've reached.

It depends on what actually broke. If you're using the ChatGPT web app or its mobile apps directly, /en/is-chatgpt-down checks the surface you're using; if you're calling the API from your own code, a product, or a tool built on top of it, this page checks the endpoint that matters to you. The two can fail independently, since they run on related but separately scaled infrastructure, so checking the wrong one can tell you everything is fine while your actual problem persists.

Not usually. Elevated latency, especially on streaming responses during periods of high demand, is a distinct pattern from a hard failure - the request eventually completes, just slower than expected, rather than erroring out or timing out completely. A genuine outage more often shows up as consistent 5xx errors or complete unresponsiveness rather than requests that are merely slow. If responses are simply taking longer across the board, that's more likely a capacity signal than a broken endpoint.